Is it possible to use only GPU and (almost) no CPU?

[Kashra]

Hi,

is there a way to say boinc to use only the GPUs but not the CPUs in the system?
It is no problem if a little percentage is used to feed the GPUs process but it should not consume all the CPU power of one or more CPUs in the system.

I have tried limiting it via the Boinc account settings profile but with no success. Boinc scales only down to one CPU and uses no GPU at all.

Thanks in advance.

[Raistmer]

Yes, there are ways to tell BOINC not to use CPUs but in first place it would be interesting why you decided to do it ?
As reminder, BOINC uses only IDLE CPU, that is, although task manager shows 100% of CPU load it doesn't mean your system will be less responsible as before.
CPU apps running on idle priority, so if CPU will be needed by another apps they will get it.

[figment]

I'd like to know if someone has an exact answer as well.  I've tried adjusting preferences in computing and resource usage sections, nothing I've tried is allowing me to use cuda processors only.  I want to see if my cuda units will process a bit faster with my core2 not doing anything.  If I don't see much of a processing time increase, then I'll run it all.

In computing preferences, I've tried:
Setting multiprocessor usage from 2 to 0 (no effect on cpu crunching)

And in Resource Share and Graphics:
Use GPU if available = YES
Use CPU = NO

Requested project update, then restarted Boinc and no luck, I'm still processing 2 CPU units and 2 cuda units.
(I'm running a Core 2 E6700 and 2 GTX260's in SLI, using Win64 Unified installer .2)

[Josef W. Segur]

A. Draconian method.  Remove all CPU apps from app_info.xml. BOINC won't fetch work for anything other than the CUDA app. Set NNT, finish, and report all CPU work before making the change, else BOINC will trash it.

B. Tricky method. In app_info, add or edit the <flops> fields. Make them about 1 million for CPU apps, for GPU app whatever billions are appropriate for the card. The Scheduler will never find work that's feasible on a 1 WMIPS CPU. This might make a good setup for rebranding, VLARs shifted to the CPU would always complete so much quicker than estimate they would have almost no effect on DCF. DISCLAIMER: I don't have a CUDA setup to try this, and don't know anyone who has. There could be some "feature" of BOINC I've overlooked which makes it impractical.
                                                                                      Joe

[efmer (fred)]

I'd like to know if someone has an exact answer as well.  I've tried adjusting preferences in computing and resource usage sections, nothing I've tried is allowing me to use cuda processors only.  I want to see if my cuda units will process a bit faster with my core2 not doing anything.  If I don't see much of a processing time increase, then I'll run it all.

In computing preferences, I've tried:
Setting multiprocessor usage from 2 to 0 (no effect on cpu crunching)

And in Resource Share and Graphics:
Use GPU if available = YES
Use CPU = NO

Requested project update, then restarted Boinc and no luck, I'm still processing 2 CPU units and 2 cuda units.
(I'm running a Core 2 E6700 and 2 GTX260's in SLI, using Win64 Unified installer .2)

Use at most 1%  CPU time is as low as it gets, but still it's pretty low impact....
Use CPU = NO means no new work, not stopping the current tasks.